Fine-scale Population Mapping And Its Correlation Analysis With Infrastructures Based On Multi-source Data

With urbanization’s rapid advancement,unequal population agglomeration and infrastructure allocation have gradually emerged in cities,which aggravates the severe vulnerability of cities to extreme natural disasters.It is a crucial way to improve people’s cognition of urban vulnerability to research the correlation analysis of population,economy,and infrastructure systems at a fine scale.However,the population and economic distribution data on the spatial scale are insufficient to support correlation analysis at a fine spatial scale.The lack of a practical analysis framework for the correlation between population and economic factors and infrastructure systems makes it a considerable challenge to accurately describe the population and economic losses caused by infrastructure cascading failures under disasters.Considering the characteristics of rich socio-economic information and spatial detail information of ubiquitous perception data,this thesis designs an efficient high-precision and advanced spatial method for population and economic factors based on the powerful hierarchical feature expression ability of the deep neural network to provide a data basis for the next step of fine analysis.This thesis further considers the relationship between disaster evolution and strong coupling between infrastructure systems,builds a dynamic model of the infrastructure system cascading failure process,and simulates the propagation mechanism of cascading failures in associated infrastructure systems.Furthermore,combining the fine-grained population and economic data to achieve progressive loss risk assessment is crucial for improving disaster emergency decisionmaking capabilities.The main research contents and innovation are summarized as follows:(1)This thesis proposes a multi-model coupled high-precision population spatialization method based on multi-source data.The distribution causes and spatial representation for the population are more complex at the fine scale.This thesis innovatively proposes a multi-scale spatial information fusion framework for modeling the natural environment and human activities based on deep learning technology.This framework solves the problem of low population estimation accuracy due to insufficient extraction of geographic attribute features and spatial features by current methods.It can integrate the rich and delicate feature information in multivariate remote sensing data and ubiquitous sensory geographic data to realize high-precision population spatialization.This model efficiently extracts rich and fine surface spatial feature information and realizes multi-scale spatial information fusion of the natural environment and human activities to achieve population mapping with 100 m spatial resolution.(2)This thesis designs an electric power consumption(EPC)spatialization estimation method by combining deep transfer learning and a random forest model.The current research only uses the luminance value of night light data to realize the spatialization of EPC,which leads to the problem of coarse-grained and low precision.This thesis innovatively proposes a fine-scale EPC spatialization method.The method uses night-light remote sensing imagery to avoid scale differences between multiple sources of geographical data and electricity consumption statistics to achieve refinement.In addition,EPC statistical survey samples are small and spatially inconsistent compared with remote sensing data.A method for extracting surface environment features based on deep transfer learning is designed.Then,the EPC spatialization method extracts the characteristic information representing the social economy,building environment,and population distribution from the corrected nighttime light imagery,high-resolution remote sensing imagery,and population density data.Finally,based on the random forest regression method,the feature set of the nature of the spatial distribution of electricity consumption is established to construct the mapping relationship between feature factors and statistics to realize the spatial feature expression of economic with a spatial resolution of about 500 m.This method can achieve accurate spatial distribution of EPC at a finescale.(3)This thesis proposes a quantitative analysis method for the socio-economic relationship of infrastructure systems in disaster scenarios.Critical infrastructure systems have a high coupling of geographic and functional relationships and complex cascading failure processes in system networks.These make it difficult to quantify the socioeconomic losses finely.From the perspective of systematics and space,this thesis takes the relationship between the infrastructure system composed of urban power system,power communication system,road traffic system,and service facility system as the research object.The proposed analysis method consists of the following: firstly,a coupled network of infrastructure systems is constructed based on their interdependencies;then,the micro-scale cascading failure model for infrastructure systems is constructed by combining the complex network theory with the disaster propagation dynamics model;finally,a fine-grained assessment of population and economic losses caused by cascading failures of infrastructure systems under disaster disturbances is achieved from a system and space perspective.The proposed fine-scale spatialization methods for population and EPC are applied in the Shenzhen area.The population with 100 m spatial resolution and the distribution of EPC with a spatial resolution of about 500 m are estimated,respectively.Based on this data,this thesis takes the "Mangosteen" typhoon disaster in Shenzhen as an example to achieve a refined quantitative analysis of social and economic losses.Specifically,the loss analysis at the micro-scale is realized through the dynamic simulation of the cascading failure process of the infrastructure coupling network.Quantitative and qualitative analysis of the results demonstrates that the proposed spatialization method can learn a non-linear mapping relationship between population and electricity consumption factors and their influencing characteristics,facilitating the spatial nonstationarity representation of the factor space.In addition,the cascading failure simulation method of the coupled network of infrastructure systems can effectively reveal the differences caused by the spatial complexity caused by the association relationship.This study has scientific significance and social value for precise urban monitoring,disaster prevention and mitigation.